Intelligent regulation and control system and method for laboratory environment

Abstract

The invention belongs to the technical field of laboratory environment regulation and control, and discloses an intelligent laboratory environment regulation and control system and method. The system comprises a parameter acquisition and correction module used for constructing and training a compensation and correction model through collected historical environmental parameters and historical compensation information corresponding to the historical environmental parameters, and performing compensation and correction on currently collected original environmental parameters based on the compensation and correction model to obtain current environmental parameters, and a target parameter determination module used for determining target parameters of the current environmental parameters. The server is used for determining target environment parameters of a laboratory according to experiment contents in the laboratory; a multi-level regulation and control framework is formed by fusing PID control, reinforcement learning, a memory table and an emergency takeover mechanism, coordinated regulation and control of parameter differences needing to be regulated corresponding to a plurality of environmental parameters and optimization of the control quantity are achieved, and it is ensured that the laboratory environment is always stabilized in a deviation threshold interval corresponding to each parameter in target environmental parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of laboratory environment regulation, and more specifically, to an intelligent laboratory environment regulation system and method. Background Art

[0002] In the scientific research field, as the core place for carrying out various experimental studies, the stability and accuracy of the environmental conditions in the laboratory play a crucial role in the accuracy and reliability of experimental results; different experimental projects have specific requirements for environmental parameters such as temperature, humidity, and gas circulation in the laboratory, and even minor fluctuations in these parameters may have a significant impact on the experimental process and results, or even lead to experimental failure. Although existing laboratory control systems have achieved automatic regulation of laboratory environmental parameters to a certain extent, most of these systems use fixed control algorithms and are inconvenient to optimize the regulation strategies of the laboratory according to changes in the laboratory environment and experimental requirements; in addition, when existing laboratory environment regulation systems regulate the environment through the collected environmental parameters, they are also prone to inaccurate regulation due to inaccurate collected environmental parameters.

[0003] In view of this, the present invention proposes an intelligent laboratory environment regulation system and method to solve the above problems. Summary of the Invention

[0004] To overcome the above defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An intelligent laboratory environment regulation system, comprising: A parameter acquisition and correction module, configured to construct and train a compensation and correction model through the collected historical environmental parameters and their corresponding historical compensation information, and compensate and correct the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters; A target parameter determination module, configured to determine the target environmental parameters of the laboratory according to the experimental content in the laboratory; A regulation and optimization module, configured to determine the difference between the parameters to be adjusted in the current environmental parameters according to the target environmental parameters, and respectively determine the control amounts of the corresponding devices through the respective PID controllers corresponding to the differences between the parameters to be adjusted, encode the differences between the parameters to be adjusted and the control amounts of the corresponding devices as state vectors and input them into the state space of reinforcement learning, select the highest reward action based on a preset memory table, optimize the control amounts of the corresponding devices using the highest reward action, and perform regulation on the devices until there is no difference between the parameters to be adjusted in the current environmental parameters.

[0005] Further, the parameter acquisition and correction module specifically includes: Preprocess the currently collected original environmental parameters in the laboratory, and the preprocessing includes outlier removal and normalization processing; Input the preprocessed original environmental parameters into the compensation and correction model to obtain the current compensation information, and combine the current compensation information with the preprocessed original environmental parameters to output the corresponding current environmental parameters.

[0006] Further, the construction and training of the compensation and correction model specifically include: Collect the historical environmental parameters in the laboratory and the historical compensation information corresponding to the historical environmental parameters, and use the historical environmental parameters and the historical compensation information corresponding to the historical environmental parameters as the sample set; Divide the sample set into a training set and a test set according to the ratio of q to 1 - q. Use the historical environmental parameters to form the input feature vector as the input layer, and use the historical compensation information corresponding to the historical environmental parameters to form the output feature vector as the output layer, so as to construct the compensation and correction model; Use the training set to train the compensation and correction model, and use the test set to test the compensation and correction model until the compensation and correction model reaches the preset evaluation index and then stop training.

[0007] Further, the target parameter determination module specifically includes: Select the experimental type corresponding to the experimental content through the system interface, determine the parameter combination corresponding to the experimental content based on the selected experimental type, and adjust the corresponding parameter combination to obtain the target environmental parameters of the laboratory.

[0008] Further, the determination of the difference in the parameters to be adjusted in the current environmental parameters according to the target environmental parameters specifically includes: Determine whether the difference between each parameter in the current environmental parameters and the target environmental parameters is within the preset corresponding deviation threshold interval; If the difference between each parameter is within the corresponding deviation threshold interval, there is no need to adjust. If the difference between each parameter is not within the corresponding deviation threshold interval, adjustment is required; Take the parameters that are not within the corresponding deviation threshold interval as the parameters to be adjusted and calculate the difference in the parameters to be adjusted. The difference in the parameters to be adjusted is obtained from the difference between the parameter corresponding to the parameter to be adjusted and the minimum adjustment value of the boundary of the corresponding deviation threshold interval.

[0009] Further, the determination of the control amount of the corresponding device through each PID controller corresponding to the difference in the parameters to be adjusted specifically includes: Comprehensively calculate the control amount of the device corresponding to the parameter to be adjusted through the proportional gain, integral gain, and derivative gain of the parameter to be adjusted to construct the expression of the corresponding PID controller; Calculate the control amount output by the PID controller corresponding to the difference in the parameters to be adjusted to the corresponding device according to the expression of the PID controller corresponding to the parameter to be adjusted.

[0010] Further, encoding the difference of the parameter to be adjusted and the control quantity of the corresponding device into a state vector and inputting it into the state space of reinforcement learning, selecting the action with the highest reward based on a preset memory table, optimizing the control quantity of the corresponding device by using the action with the highest reward, and performing regulation on the device until there is no difference in the parameter to be adjusted in the current environmental parameters, specifically including: Setting a period, encoding the difference of the parameter to be adjusted and the control quantity of the corresponding device into a state vector according to the period and inputting it into the state space of reinforcement learning; Then, the state vector selects an action from a preset memory table. The action selection includes: retrieving the high-reward action of the most similar state from the memory table as the action. If there is no matching item in the memory table, use the ε-greedy strategy to explore a new action; Adding the adjustment amount of the control quantity of the device corresponding to the selected action to the control quantity of the corresponding device output by the PID controller to form a final control signal and sending it to the corresponding device for execution. Calculating the reward for the action based on the regulation time, regulation times, and safety violation times corresponding to the selected action to obtain a reward value. Inserting the action corresponding to the reward value and the state vector into the memory table and updating the memory table.

[0011] Further, the memory table specifically includes: Combining the action corresponding to the state vector and the reward value into an array and storing it in the memory table as a new policy, setting an update rule, and updating the memory table based on the update rule.

[0012] Further, the regulation optimization module further includes an emergency takeover regulation mechanism. The emergency takeover regulation mechanism specifically includes: When any parameter exceeds the corresponding preset safety threshold, switching the device corresponding to the parameter greater than or equal to the preset safety threshold to the maximum allowable regulation mode until the corresponding parameter is less than the preset safety threshold.

[0013] The present invention provides another technical solution: an intelligent regulation method for a laboratory environment, including the following steps: Step 1: Constructing and training a compensation and correction model through the collected historical environmental parameters and their corresponding historical compensation information, and compensating and correcting the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters; Step 2: Determining the target environmental parameters of the laboratory according to the experimental content in the laboratory; Step 3: Determine the difference in adjustable parameters in the current environmental parameters according to the target environmental parameters, and use each PID controller corresponding to the difference in adjustable parameters to determine the control amount of the corresponding device. Encode the difference in adjustable parameters and the control amount of the corresponding device as a state vector and input it into the state space of reinforcement learning. Select the highest reward action based on the preset memory table, use the highest reward action to optimize the control amount of the corresponding device, and perform regulation and control on the device until there is no difference in adjustable parameters in the current environmental parameters.

[0014] Technical effects and advantages of an intelligent laboratory environment regulation system and method of the present invention: 1. By compensating and correcting the currently collected environmental parameters through the compensation and correction model, the error existing in the original environmental parameters caused by sensor noise and environmental interference can be solved, making the current environmental parameters relatively accurate, providing a basis for subsequent precise regulation; 2. Each difference in adjustable parameters corresponding to a parameter corresponds to an independent PID controller, and multiple PID controllers respectively perform precise control on different parameters, realizing high-precision collaborative regulation and optimization of the differences in adjustable parameters corresponding to multiple environmental parameters. Among them, when performing PID control and reinforcement learning, first use the corresponding PID controller to determine the control amount of the corresponding device according to the difference in adjustable parameters, and then use reinforcement learning to select the highest reward action based on the preset memory table to optimize the control amount. This can not only quickly perform preliminary regulation on the difference in adjustable parameters corresponding to the current environmental parameters, but also continuously learn and optimize to find the optimal regulation strategy, improving the regulation efficiency and accuracy of the corresponding PID controller until the current environmental parameters meet the requirements of the target environmental parameters; 3. By combining the actions and reward values corresponding to the state vector into an array and storing it as a new policy in the memory table, the memory table provides historical data support with higher reward values for reinforcement learning, helps the system better understand the relationship between states and actions, can accelerate the learning process, and improve the optimization efficiency; The emergency takeover regulation mechanism provides reliable safety protection for the laboratory environment, can effectively avoid damage to experimental equipment, experimental samples, and even personnel caused by abnormal environmental parameters, and ensure the safe operation of the laboratory. Description of the Drawings

[0015] Figure 1 It is a schematic structural diagram of an intelligent laboratory environment regulation system of the present invention; Figure 2 It is a schematic flowchart of an intelligent laboratory environment regulation method of the present invention. Detailed Embodiment

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] Although the existing automated control systems for laboratory environmental parameters have achieved automatic regulation to a certain extent, most of these automated control systems adopt fixed control algorithms and are difficult to flexibly adjust and coordinately regulate according to the changes in the laboratory environment and the diversity of experimental requirements. For example, there are often environmental dynamic changes such as multivariable coupling and non-linear environments in existing laboratories. How to achieve parallel regulation of multiple environmental parameters and optimize the control quantity to improve the control effect in complex scenarios is a difficult problem that urgently needs to be solved. In addition, the original environmental parameters collected in existing laboratories also have the problem of inaccurate collected parameters due to the service life of sensors, electromagnetic radiation, and the influence of other environmental parameters. If the directly collected original environmental parameters are used for regulation, it is easy to lead to inaccurate regulation results. Based on this, the present invention proposes a laboratory environment intelligent regulation system and method to solve the above problems.

[0018] Please refer to Figure 1 As shown, the intelligent laboratory environment regulation system in this embodiment includes a parameter acquisition and correction module, a target parameter determination module, and an intelligent regulation and optimization module. The above-mentioned parameter acquisition and correction module is used to construct and train a compensation and correction model through the collected historical environmental parameters and their corresponding historical compensation information, and compensate and correct the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters. It should be noted that compensating and correcting the currently collected environmental parameters through the compensation and correction model can solve the errors existing in the original environmental parameters caused by sensor noise and environmental interference, making the current environmental parameters relatively accurate and providing a basis for subsequent precise regulation. The above-mentioned target parameter determination module is used to determine the target environmental parameters of the laboratory according to the experimental content in the laboratory. It should be noted that determining the target environmental parameters through the experimental content can ensure that the target environmental parameters of the laboratory correspond to the experimental content, making the parameters in the laboratory more in line with the specific environmental requirements of the experimental content in subsequent intelligent regulation, and ensuring the accuracy and reliability of parameter regulation in the laboratory. The above intelligent regulation and optimization module is used to determine the difference in parameters to be adjusted in the current environmental parameters according to the target environmental parameters, and respectively determine the control amounts of the corresponding devices through each PID controller corresponding to the difference in parameters to be adjusted. The difference in parameters to be adjusted and the control amounts of the corresponding devices are encoded as a state vector and input into the state space of reinforcement learning. Based on a preset memory table, the highest reward action is selected, and the control amounts of the corresponding devices are optimized using the highest reward action, and the devices are regulated until there is no difference in parameters to be adjusted in the current environmental parameters. It should be noted that when performing PID control and reinforcement learning, first, the control amounts of the corresponding devices are determined by the corresponding PID controllers according to the difference in parameters to be adjusted, and then the reinforcement learning is used to select the highest reward action based on the preset memory table to optimize the control amounts. This can not only quickly perform preliminary regulation on the difference in parameters to be adjusted corresponding to the current environmental parameters, but also, through continuous learning and optimization, find the optimal regulation strategy, improve the regulation efficiency and accuracy of the corresponding PID controllers until the current environmental parameters meet the requirements of the target environmental parameters; In this embodiment, each difference in parameters to be adjusted corresponding to a parameter corresponds to an independent PID controller, and multiple PID controllers respectively perform precise control on different parameters, realizing high-precision collaborative regulation and optimization of the differences in parameters to be adjusted corresponding to multiple environmental parameters. Among them, when performing PID control and reinforcement learning, first, the control amounts of the corresponding devices are determined by the corresponding PID controllers according to the difference in parameters to be adjusted, and then the reinforcement learning is used to select the highest reward action based on the preset memory table to optimize the control amounts. This can not only quickly perform preliminary regulation on the difference in parameters to be adjusted corresponding to the current environmental parameters, but also, through continuous learning and optimization, find the optimal regulation strategy, improve the regulation efficiency and accuracy of the corresponding PID controllers until the current environmental parameters meet the requirements of the target environmental parameters; By combining the actions and reward values corresponding to the state vector into an array and storing it as a new policy in the memory table, the memory table provides historical data support with higher reward values for reinforcement learning, helps the system better understand the relationship between states and actions, can accelerate the learning process, and improve the optimization efficiency; The emergency takeover regulation mechanism provides reliable safety protection for the laboratory environment, can effectively avoid damage to experimental equipment, experimental samples, and even personnel caused by abnormal environmental parameters, and ensure the safe operation of the laboratory.

[0019] As an optional embodiment: The parameter acquisition and correction module specifically includes: Perform preprocessing on the original environmental parameters currently collected in the laboratory, and the preprocessing includes outlier removal and normalization processing; Input the preprocessed original environmental parameters into the compensation and correction model to obtain the current compensation information, and combine the current compensation information with the preprocessed original environmental parameters to output the corresponding current environmental parameters; It should be noted that, for example, if the original environmental parameters in the laboratory include temperature, humidity, and gas flow rate, the current compensation information includes temperature compensation value, humidity compensation value, and gas flow rate compensation value; outlier rejection includes cases where temperature, humidity, and gas flow rate exceed the historical mean ± 3 times the standard deviation (i.e., the 3σ principle); the original environmental parameters after outlier rejection are normalized using Min-Max; furthermore, by rejecting outliers, unreasonable data caused by sensor failures, measurement errors, or sudden interferences can be effectively removed, avoiding the misguidance of abnormal data on subsequent control and decision-making, ensuring the quality and reliability of the input data, and thus providing a basic guarantee for the stable operation of the system; by using Min-Max to convert the original environmental parameters after outlier rejection to a unified range (usually [0,1]), the differences in the dimensions and numerical ranges of different original environmental parameters are eliminated, which can improve the training efficiency and accuracy of the compensation correction model and avoid calculation errors or model deviations caused by overly large numerical differences; the preprocessed original environmental parameters are input into the compensation correction model to obtain the current compensation information (such as temperature compensation value, humidity compensation value, and gas flow rate compensation value), and it is combined with the original environmental parameters (the corresponding parameters are added or subtracted from the compensation information) to output more accurate current environmental parameters; the compensation correction model can dynamically adjust and correct the original measurement data according to historical data and known environmental characteristics, compensating for the deviations caused by factors such as sensor measurement errors, environmental interferences, or imperfect models, making the current environmental parameters relatively accurate and providing a basis for subsequent precise regulation.

[0020] As an optional embodiment: The constructing and training of the compensation correction model specifically includes: Collect historical environmental parameters in the laboratory and the corresponding historical compensation information for the historical environmental parameters, and use the historical environmental parameters and the corresponding historical compensation information as a sample set; Divide the sample set into a training set and a test set according to the ratio of q to 1 - q. Use the historical environmental parameters to form an input feature vector as the input layer, and use the corresponding historical compensation information for the historical environmental parameters to form an output feature vector as the output layer, thereby constructing the compensation correction model; Use the training set to train the compensation correction model, and use the test set to test the compensation correction model until the compensation correction model reaches the preset evaluation index and then stop training; Specifically, collect 1000 sets of historical environmental parameters (temperature, humidity, air circulation) in the laboratory and their corresponding historical compensation information (such as temperature compensation value, humidity compensation value, air circulation compensation value), preprocess the historical environmental parameters and historical compensation information, including outlier removal and normalization, and take each set of historical environmental parameters and their corresponding historical compensation information as a sample; divide the dataset in an 8:2 ratio to obtain 800 sets of training data and 200 sets of test data. Use the historical environmental parameters as the input feature vector X = [temperature, humidity, air circulation] and the historical compensation information as the output feature vector Y = [temperature compensation value, humidity compensation value, air circulation compensation value]. Use a multi-layer perceptron (MLP) to build the network architecture of the compensation correction model. Among them, the input layer has 3 neurons (corresponding to 3 environmental parameters respectively), the hidden layer has 2 layers, with 10 neurons in each layer, the activation function is ReLU, the output layer has 3 neurons (corresponding to 3 compensation values respectively), and there is no activation function. Initialize the biases and weights of the compensation correction model in the range of [-0.1, 0.1]; use the training set to train the compensation correction model, with the mean square error minimization as the loss function of the training set. The calculation formula of the mean square error is: , where is the actual temperature compensation value in the i-th sample and the predicted temperature compensation value is the square of the error, is the actual humidity compensation value in the i-th sample and the predicted humidity compensation value is the square of the error, is the actual air circulation compensation value in the i-th sample and the predicted air circulation compensation value is the square of the error, is the number of samples in the dataset, and i is the index of the samples in the dataset; then, use stochastic gradient descent to update the model parameters (weights and biases) along the negative gradient direction of the loss function, gradually approaching the minimum value of the loss function. Set the batch size for each iteration to 32 and the maximum number of iterations to 100. If the MSE of the test set has not been lower than the historical MSE in 10 consecutive rounds of iteration, stop the training, that is, complete the construction and training of the compensation correction model. In addition, the compensation correction model can be further optimized: for example, deploy a sliding window mechanism (such as window size N = 200), only retain the latest 200 sets of environmental parameters and compensation information as training samples, regularly (such as every 24 hours) add newly collected data to the sliding window and remove old data, so that the compensation correction model continuously adapts to dynamic factors such as sensor aging and environmental changes. Based on the initial model, perform incremental training every 24 hours using the newly added data in the sliding window, fine-tune the parameters of the compensation correction model with a small learning rate (such as 10% of the initial learning rate), and retain 10% of the window data as the validation set during training (to prevent the model from forgetting important early patterns). If the validation set loss increases continuously for 3 times, terminate the training to prevent overfitting, so as to complete the online learning of the compensation correction model, make the compensation correction model keep up with sensor aging and environmental changes, and ensure accurate compensation correction; It should be noted that regarding the collection of historical compensation information, the first method is to long-term monitor the environmental parameters in the laboratory, analyze the deviation between the sensor measurement values and the actual environmental conditions in the historical data, and calculate through statistical methods (such as mean, median, standard deviation, etc.), and use this as the compensation information. For example, in the monitoring data of the humidity sensor in the laboratory in the past year, the average deviation between the humidity measurement value and the actual humidity is -2%, then the humidity compensation value can be set to +2%; the second method is to establish an environmental model (such as a physical model or a data-driven model), predict the changes in environmental parameters, and calculate the deviation between the theoretical value and the actual measurement value, and these deviations can be used as compensation information; for example, the predicted value of the gas flow rate model in the laboratory is 0.3 m / s, while the sensor measurement value is 0.28 m / s, then the gas flow rate compensation value is +0.02 m / s; furthermore, through the collection and application of historical compensation information, the sensor measurement error can be effectively corrected, the accuracy of environmental parameters can be improved, and it can be made closer to the true value or the target value; by collecting the historical environmental parameters in the laboratory and their corresponding historical compensation information as a sample set, the compensation correction model can learn the internal relationship between environmental parameters and compensation information, which enables the compensation correction model to accurately predict the required compensation value according to the current environmental parameters, thereby correcting the error in the original measurement value and improving the accuracy of environmental parameters, solving the problem that the sensor may be affected by various factors during long-term operation, such as equipment aging, environmental interference, etc., resulting in measurement errors; thus, the compensation correction model can dynamically adjust the compensation information according to the original environmental parameters to ensure the stability and accuracy of the current environmental parameters; through the compensation correction model, the parameter fluctuations caused by sensor errors or environmental interference can be reduced, and the current environmental parameters can be made more accurate, providing a basis for subsequent precise control.

[0021] As an optional embodiment: The target parameter determination module specifically includes: Select the experimental type corresponding to the experimental content through the system interface, determine the parameter combination corresponding to the experimental content based on the selected experimental type, and adjust the corresponding parameter combination to obtain the target environmental parameters of the laboratory; It should be noted that the system interface is generally displayed through display devices such as display screens; there are multiple experiment types in the system, and each experiment type is preset with corresponding parameter combinations. For example, if the experimental content belongs to the chemical experiment type, the parameter combination can be (temperature, relative humidity, air flow velocity); adjust the corresponding parameter combination: for example, the values of the parameters in the parameter combination of experiment type A are (temperature 25°C ± 1°C, relative humidity 45% ± 2%, air flow velocity 0.1 - 0.2 m / s), and relevant experimental personnel adjust the parameters and their corresponding values in the parameter combination of the above experiment type A according to the specific requirements of the experimental content, such as (temperature 29°C ± 1°C, relative humidity 60% ± 3%, air flow velocity 0.4 - 0.6 m / s) to obtain the target environmental parameters corresponding to the experimental content; determine the target environmental parameters through the experimental content, so as to ensure that the target environmental parameters of the laboratory correspond to the experimental content, making the parameters of the laboratory more in line with the specific environmental requirements of the experimental content in subsequent intelligent regulation, and ensuring the accuracy and reliability of the parameter regulation in the laboratory.

[0022] As an optional embodiment: determining the difference between the parameters to be adjusted in the current environmental parameters according to the target environmental parameters specifically includes: Determine whether the difference between each parameter in the current environmental parameters and the target environmental parameters is within the preset corresponding deviation threshold range; If the difference between each parameter is within the corresponding deviation threshold range, no adjustment is required; if the difference between each parameter is not within the corresponding deviation threshold range, adjustment is required; Take the parameter that is not within the corresponding deviation threshold range as the parameter to be adjusted and calculate the difference between the parameters to be adjusted. The difference between the parameters to be adjusted is obtained by the difference between the parameter corresponding to the parameter to be adjusted and the minimum adjustment value of the boundary of the corresponding deviation threshold range.

[0023] It should be noted that the deviation threshold range is generally set according to the accuracy requirements of the actual task for environmental parameters to allow the maximum deviation range. For example, in a laboratory for high-precision experiments (such as chemical reaction control), the temperature deviation needs to be strictly controlled within ±0.5°C, so the corresponding deviation threshold range for temperature is set as [-0.5°C, +0.5°C]. If it is only for general environmental monitoring, the corresponding deviation threshold range for temperature can be relaxed to ±2°C, and the range is set as [-2°C, +2°C]. If the current temperature is 22°C and the target temperature is 17°C, and their difference is not within the corresponding deviation threshold range [-1°C, +1°C], the current temperature is taken as the parameter to be adjusted, and the current temperature is regulated. When regulating the current temperature, the difference of the parameter to be adjusted is 17°C - 22°C ± 1°C = -4°C or -6°C. Taking the minimum adjustment value, that is, -4°C is the difference of the parameter to be adjusted. That is to say, the current temperature needs to be regulated to decrease by at least 4°C so that the difference between the regulated current temperature and the target temperature is within the corresponding deviation threshold range. Furthermore, by determining whether the difference between each parameter in the current environmental parameters and the target environmental parameters is within the preset corresponding deviation threshold range, it can be determined whether the current environmental parameters need to be regulated. Through the setting of the deviation threshold range corresponding to each parameter, each parameter can fluctuate within the allowable range and reduce ineffective operations.

[0024] As an optional embodiment: The step of respectively determining the control amount of the corresponding device through the difference of the parameter to be adjusted by each PID controller specifically includes: Comprehensively calculating the control amount of the device corresponding to the parameter to be adjusted through the proportional gain, integral gain, and derivative gain of the parameter to be adjusted to construct the expression of the corresponding PID controller; Calculating the control amount output by the PID controller corresponding to the difference of the parameter to be adjusted to the corresponding device according to the expression of the PID controller corresponding to the parameter to be adjusted. The PID controller expression is: , In the formula, represents the control amount (such as the power of the heater / cooler, the power of the humidifier, the power of the exhaust fan) output by the PID controller to the corresponding device, and is used to output to the device (heater / cooler, humidifier, exhaust fan) to regulate the parameter to be adjusted; represents the proportional gain of the parameter to be adjusted, which is used to determine the response intensity of the PID controller to the current error of the parameter to be adjusted; represents the error information between the target value and the actual measured value of the parameter to be adjusted; represents the integral gain of the parameter to be adjusted, which is used to determine the response intensity of the PID controller to the past cumulative error of the parameter to be adjusted to eliminate the steady-state error; represents the integral part of the error corresponding to the parameter to be adjusted, which represents the cumulative value of the error of the parameter to be adjusted from the initial moment to the current moment; The differential gain representing the parameter to be adjusted is used to determine the response intensity of the PID controller to the rate of change of the error of the parameter to be adjusted, so as to predict the future trend of the error of the parameter to be adjusted and make an early adjustment; It represents the differential term of the error corresponding to the parameter to be adjusted, indicating the rate of change of the error of the parameter to be adjusted; It should be noted that in the PID controller corresponding to temperature, the determination process of Kp, Ki, and Kd is as follows: Establish a laboratory temperature data model, a first-order system: , where represents the transfer function of the temperature system, represents the gain of the temperature system, represents the time constant. For example: Collect the temperature change curve over time of the initial temperature in the laboratory environment from 20°C to the steady state of 35°C and the heater power step from 0% to 50%, perform data fitting, and obtain K = (35 - 20) / 50 = 0.3°C, and the time constant Ts ≈ 400 seconds (the time required for the heater response to reach 63.2%); Substitute into the Ziegler-Nichols formula to calculate the PID parameters corresponding to the temperature controller: Kp = 0.9Ts / K = 0.9 * 400 / 0.3 = 1200, Ki = 0.27Ts / K 2 = 0.27 * 400 / 0.3 2 = 1200, Kd = 0.3Ts 2 / K = 0.3 * 400 2 / 0.3 = 160000. Usually, multiply the above values of Kp and Ki by the safety factor 0.6, that is, the values of Kp and Ki of the temperature corresponding controller are 720 and 720 respectively, and multiply the value of Kd by the safety factor 0.01, that is, the value of Kd of the temperature corresponding controller is 1600; In the PID controller corresponding to humidity, the determination process of Kp, Ki, and Kd is as follows: Establish a laboratory humidity data model, a first-order system with lag: , represents the transfer function of the humidity system, represents the gain of the humidity system, represents the time constant, is the Laplace expression of the pure lag link, indicating that the signal is delayed in time by seconds, represents the lag time. For example: Collect the humidity change curve over time of the initial humidity in the laboratory environment from 40%RH to the steady state of 65%RH and the humidifier power step from 0% to 30%, perform data fitting, and obtain K = (65 - 40) / 30 = 0.833, the time constant Ts ≈ 500 seconds, and the lag time seconds. For the system with lag, use the improved Ziegler-Nichols formula to calculate the PID parameters corresponding to the humidity controller: , , ; that is, the values of Kp, Ki, and Kd of the humidity corresponding controller are 506, 184, and 16206 respectively; In the PID controller corresponding to the air circulation rate, the determination process of Kp, Ki, and Kd is as follows: Establish a data model of the air circulation rate in the laboratory, a second-order system: , where represents the transfer function of the air circulation rate system, represents the gain of the air circulation rate system, represents the natural frequency, represents the complex variable in the Laplace transform, which is used to describe the dynamic characteristics of the air circulation rate in the complex frequency domain, represents the damping ratio. For example: Collect the power of the exhaust fan in the laboratory environment stepping from 0% to 40%, record the change in wind speed, the wind speed rises from 0 m / s to 3 m / s, the rise time is about 40 seconds, there is overshoot, perform data fitting, obtain K = 3 / 40 = 0.075 m / s, and estimate to get , ; Use the improved Ziegler-Nichols formula to calculate the PID parameters corresponding to the air circulation rate controller: , , , that is, the values of Kp, Ki, and Kd of the air circulation rate corresponding controller are 2.67, 0.53, and 13.33 respectively; If the parameter to be adjusted is the temperature, the target temperature is 25 °C, the actual measured temperature is 28 °C, and the corresponding deviation threshold interval is ±1 °C, then the temperature difference of the parameter to be adjusted is -2 °C; The proportional gain determines the response intensity of the PID controller to the current temperature error, the integral gain can eliminate the historical cumulative error of the temperature (such as the long-term steady-state deviation of the temperature), and the derivative gain can predict the error change trend of the temperature. represents the final control quantity of the device corresponding to the parameter to be adjusted (such as an air conditioner cooler). The system converts the final control quantity into a control signal and outputs it to the corresponding device to regulate the temperature in the laboratory. For example, the air conditioner receives u(t) = +2.5, indicating that the refrigeration power needs to be increased by 2.5 units; Through the PID controllers corresponding to the above parameters to be adjusted, the system can automatically monitor the current environmental parameters in the laboratory and adjust the operating state of the device according to the deviation situation, realizing precise control of each parameter in the current environmental parameters and greatly improving the control efficiency.

[0025] As an optional embodiment: encoding the difference between the parameters to be adjusted and the control quantity of the corresponding device into a state vector and inputting it into the state space of reinforcement learning, and selecting the action with the highest reward based on a preset memory table, using the action with the highest reward to optimize the control quantity of the corresponding device and performing regulation on the device until there is no difference in the parameters to be adjusted in the current environmental parameters, specifically including: Set a period, and encode the difference between the parameters to be adjusted and the control quantity of the corresponding device into a state vector according to the period and input it into the state space of reinforcement learning. The state vector is expressed as: , where represents the state vector, represents the m-th difference in the parameters to be adjusted in the current period, m is the index of the difference in the parameters to be adjusted, represents the control quantity of the corresponding device for the m-th difference in the parameters to be adjusted in the previous period, t represents the current period, and t - 1 represents the previous period; Then, action selection is performed on the state vector from the preset memory table. The action selection includes: retrieving the high-reward action of the most similar state from the memory table as the action , if there is no matching item in the memory table, use the ε-greedy strategy to explore new actions; Superimpose the adjustment amount of the control quantity of the corresponding device for the selected action on the control quantity of the corresponding device output by the PID controller to form a final control signal and send it to the corresponding device for execution. Calculate the reward for the action based on the regulation time, regulation times, and safety violation times corresponding to the selected action to obtain the reward value. Insert the action corresponding to the reward value and the state vector into the memory table and update the memory table , the expression for reward calculation is: , where represents the reward value, represents the absolute value of the m-th difference in the parameters to be adjusted (such as the absolute value of the temperature deviation), M represents the number of differences in the parameters to be adjusted, represents 0 when all differences in the parameters to be adjusted are within the corresponding deviation threshold range, otherwise 1; TK represents the regulation time, which is the duration from execution to completion for a single regulation; OS represents the regulation times, which is the total number of regulation commands issued by the system to the device within a period; SV represents the number of safety violations; a1, a2, and a3 respectively represent the weight values of the regulation time, regulation times, and safety violation times, and their value ranges are in [0, 1], The value is greater than or equal to 1. Among them, in experiments with high requirements for regulation efficiency and rapid changes in indoor temperature: a1 increases, while a2 and a3 decrease; in long-term experiments with high requirements for stability: a2 increases, and a1 and a3 are in balance; in high-risk laboratories: a3 increases, and a1 and a2 decrease; Specifically, the period should be set according to the specific experimental content. For example, for experimental content with high environmental requirements (such as chip manufacturing), a period of 1 to 10 seconds is selected; for ordinary laboratories, a period of 1 to 10 minutes is selected. , where represents the adjustment amount of the device control quantity corresponding to the difference in the m-th parameter to be adjusted (such as temperature "-1°C", "power +5%"); taking a chemical laboratory as an example, it includes three parameters to be adjusted, namely temperature, humidity, and air circulation, as well as three differences in the parameters to be adjusted. Each parameter to be adjusted has a corresponding PID controller; initialize the initial parameters of the temperature, humidity, and wind speed PID controllers; construct a state vector, including the differences in the parameters to be adjusted (ΔC1, ΔC2, ΔC3) in the current period and the control quantities (uC1, uC2, uC3) of the corresponding devices in the previous period; set the upper limit of the memory table capacity to 500 entries, and define the similarity matching method (such as cosine similarity) and update rule of the memory table; It should be noted that the state vector only contains the difference of the parameters to be adjusted and the control amount of the corresponding equipment, and the action only contains the adjustment amount of the control amount. In order to avoid the control amount exceeding the safety range of the equipment, the upper and lower limits of the action adjustment amount must be clearly defined: Temperature control: The adjustment amount is limited to ±5% (such as the power change of the heater / cooler does not exceed ±5% of the rated power), with a step size of 0.5%, to ensure that the temperature control is stable and within the tolerance range of the equipment; Humidity control: The adjustment amount is limited to ±10% (the power change of the humidifier / dehumidifier does not exceed ±10% of the rated power) , step size 1%, balance humidity regulation efficiency and equipment safety; air circulation control: the adjustment amount is limited to ±15% (the power change of the exhaust fan does not exceed ±15% of the rated speed), step size 1%, to avoid equipment overload caused by sudden changes in wind speed; then, the specific execution steps of each cycle (S10~S13) are as follows: S10 constructs a state vector: calculates the difference in the parameters to be adjusted and the control amount of the corresponding equipment, and constructs a state vector with the difference in the parameters to be adjusted and the control amount of the corresponding equipment, such as [+2℃, 23%, -3%RH, 45%, -0.02 m / s, 30%]; S11 Action selection: retrieve the top 10% strategies with the highest similarity to the state vector in the memory table (calculated using cosine similarity. If a similar strategy is found, use the corresponding high-reward action (such as "ΔuC1_temperature=+2%, ΔuC2_humidity=-3%, ΔuC3_excessive air circulation=+5%). If there is no match, randomly select an action with a probability of ε=0.5 (such as trying "ΔuC1__temperature=+1.8%"), and add the adjustment amount of the device control amount corresponding to the selected action to the PID control. The controller outputs the control amount of the corresponding device to form the final control signal and send it to the corresponding device for execution; S12 reward calculation: calculate the reward value of the selected action according to the above reward calculation formula; S13 memory table update: insert the (St, At, Rt) corresponding to the new strategy into the memory table; sort all strategies in the memory table in descending order by Rt, retain the top 20% high reward strategies (such as strategies with Rt≥-0.5), and eliminate the strategies corresponding to the last 80% low rewards; if the memory table reaches the upper limit of 500 entries, delete the 20% strategies with the lowest Rt; Furthermore, by periodically optimizing the control variables of the devices corresponding to multiple PID controllers through reinforcement learning, the output of the control variables can be optimized according to environmental parameters, thereby achieving precise control of adjustable parameters such as temperature, humidity, and air circulation, enabling each adjustable parameter to quickly and stably reach the deviation threshold range allowed by the set target environmental parameters, and avoiding the problems of adjustment lag or excessive oscillation that may occur in traditional fixed control variable control; through reinforcement learning, the optimal action can be selected or new actions can be explored based on the current state vector to superimpose adjustment variables on the control variables, enabling the system to adapt to changes in the laboratory environment; through the calculation of reward values, the selection of different actions can be evaluated and optimized. Among them, the calculation of reward values comprehensively considers multiple factors such as the difference in adjustable parameters, regulation time, regulation times, and safety violation times, enabling the system to not only pursue rapid and stable parameters but also take into account the efficient use of resources and regulation efficiency; for example, during the regulation process, the system will preferentially select those strategies that can stabilize the parameters within the target range in a shorter time and with fewer regulation times, avoiding unnecessary resource waste and frequent regulation operations, and improving the overall operating efficiency of the system.

[0026] As an optional embodiment: The memory table for storing actions and control variables specifically includes: Merge the action corresponding to the state vector and the reward value into an array and store it as a new strategy in the memory table, set an update rule, and update the memory table based on the update rule; The update rule includes: Insert the new strategy into the memory table and sort it in descending order according to Rt; retain the top 20% of the high-reward strategies and eliminate the bottom 80%; if the memory table is full (such as a capacity of 1000 entries), eliminate the strategy with the lowest reward; It should be noted that the state vector of the current period is stored in the form of an array, and the cosine similarity algorithm is used to calculate the similarity between state vectors. The similarity threshold is set to 0.9 (that is, only when the similarity is greater than 90% can it be adopted); capacity management: when the number of strategies in the memory table reaches 500, delete them according to the following priority: retain all strategies; if still need to delete, then eliminate the lowest strategy; aging mechanism: each record is attached with a timestamp. If a certain strategy has not been accessed for more than 96 hours, its weight will be automatically reduced (reducing its retrieval probability). Furthermore, the memory table can provide historical data support with higher reward values for reinforcement learning, helping the system better understand the relationship between states and actions, accelerating the learning process, and improving the optimization efficiency.

[0027] As an optional embodiment: The regulation optimization module further includes an emergency takeover regulation mechanism, and the emergency takeover regulation mechanism specifically includes: When any parameter exceeds the corresponding preset safety threshold, the device corresponding to the parameter greater than or equal to the preset safety threshold is switched to the maximum allowable regulation mode until the corresponding parameter is less than the preset safety threshold; It should be noted that while any parameter exceeds the corresponding preset safety threshold, the control of reinforcement learning is paused and the current memory table is retained. When all parameters are less than the corresponding preset safety threshold, the reinforcement learning is reactivated; for example, in a chemical laboratory, if a reaction involving chlorine-containing reagents is in progress and the safety threshold for the safe chlorine concentration is 1 mg / m³ (about 0.3 ppm), when the relevant sensor detects that the chlorine concentration in the laboratory reaches 1 mg / m³, the exhaust fan power in the maximum regulation mode (i.e., the maximum control amount) is immediately started, and at the same time, the control of reinforcement learning is paused and the current memory table is retained to ensure that the emergency handling process will not be interfered by the dynamic adjustment of reinforcement learning in case of an emergency; at this time, the system continuously monitors the chlorine concentration. Once the chlorine concentration drops below the safety threshold (such as 0.5 mg / m³), the system automatically adjusts the power of the exhaust fan to the normal working mode (such as the power is restored to the normal mode), and reactivates the reinforcement learning to continue the optimization control; furthermore, the emergency takeover regulation mechanism provides reliable safety protection for the laboratory environment, can effectively avoid damage to experimental equipment, experimental samples and even personnel due to abnormal environmental parameters, and ensure the safe operation of the laboratory.

[0028] Please refer to Figure 2 As shown, the intelligent regulation method for a laboratory environment described in this embodiment includes the following steps: Step 1: Construct and train a compensation and correction model based on the collected historical environmental parameters and their corresponding historical compensation information, and compensate and correct the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters; Step 2: Determine the target environmental parameters of the laboratory according to the experimental content in the laboratory; Step 3: Determine the difference between the parameters to be adjusted in the current environmental parameters according to the target environmental parameters, and respectively determine the control amounts of the corresponding devices through the respective PID controllers corresponding to the differences between the parameters to be adjusted. Encode the difference between the parameters to be adjusted and the control amounts of the corresponding devices as a state vector and input it into the state space of the reinforcement learning. Select the highest reward action based on the preset memory table, use the highest reward action to optimize the control amounts of the corresponding devices and perform regulation on the devices until there is no difference between the parameters to be adjusted in the current environmental parameters; In this embodiment, through the settings of PID control, reinforcement learning, memory table, and emergency takeover control mechanism in the control strategy, the collaborative work of multiple PID controls can be used to execute control tasks in real time and precisely, ensuring the stability of environmental parameters during the control process. Reinforcement learning can continuously learn and adjust the control amount of the PID controller corresponding to the difference in the parameters to be adjusted, enabling the system to better adapt to different environmental changes and experimental requirements. Over time, it can more efficiently achieve more accurate and higher-efficiency control amounts. By learning and making decisions based on the historical data in the memory table, repeated exploration and errors can be avoided, accelerating convergence to the optimal strategy. As the last line of defense in the control strategy, the emergency takeover control mechanism can intervene in a timely manner and take preset emergency measures when abnormal situations occur in the laboratory, ensuring that the environmental parameters of the laboratory do not exceed the safe range, guaranteeing the smooth progress of the experiment and the safety of the equipment, and improving the reliability and safety of the system.

[0029] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present invention can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0030] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one type, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0031] As mentioned above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

[0032] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included in the protection scope of the present invention.

Claims

1. An intelligent laboratory environment control system, characterized in that, Including: A parameter acquisition and correction module, which is used to construct and train a compensation and correction model based on the collected historical environmental parameters and their corresponding historical compensation information, and compensate and correct the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters; A target parameter determination module, which is used to determine the target environmental parameters of the laboratory according to the experimental content in the laboratory; A regulation and optimization module, which is used to determine the difference of the parameters to be adjusted in the current environmental parameters according to the target environmental parameters, and respectively determine the control quantities of the corresponding devices through each PID controller corresponding to the difference of the parameters to be adjusted, encode the difference of the parameters to be adjusted and the control quantities of the corresponding devices as a state vector and input it into the state space of reinforcement learning, select the highest reward action based on a preset memory table, use the highest reward action to optimize the control quantities of the corresponding devices and perform regulation on the devices until there is no difference in the parameters to be adjusted in the current environmental parameters.

2. The intelligent control system for laboratory environment according to claim 1, wherein The parameter acquisition and correction module specifically includes: Preprocess the currently collected original environmental parameters in the laboratory, and the preprocessing includes outlier removal and normalization processing; Input the preprocessed original environmental parameters into the compensation and correction model to obtain the current compensation information, and combine the current compensation information with the preprocessed original environmental parameters to output the corresponding current environmental parameters.

3. The intelligent laboratory environment regulation system according to claim 2, wherein, The construction and training of the compensation and correction model specifically includes: Collect the historical environmental parameters in the laboratory and the historical compensation information corresponding to the historical environmental parameters, and use the historical environmental parameters and the historical compensation information corresponding to the historical environmental parameters as a sample set; Divide the sample set into a training set and a test set according to the ratio of q to 1-q, use the historical environmental parameters to form an input feature vector and serve as the input layer, use the historical compensation information corresponding to the historical environmental parameters to form an output feature vector and serve as the output layer, and thus construct a compensation and correction model; Use the training set to train the compensation and correction model, and use the test set to test the compensation and correction model until the compensation and correction model reaches the preset evaluation index and then stop training.

4. An intelligent control system for a laboratory environment according to claim 1, characterized in that, The target parameter determination module specifically includes: Select the experimental type corresponding to the experimental content through the system interface, determine the parameter combination corresponding to the experimental content based on the selected experimental type, and adjust the corresponding parameter combination to obtain the target environmental parameters of the laboratory.

5. An intelligent laboratory environment control system according to claim 1, characterized in that, The determination of the difference of the parameters to be adjusted in the current environmental parameters according to the target environmental parameters specifically includes: Determine whether the difference of each parameter between the current environmental parameters and the target environmental parameters is within the preset corresponding deviation threshold interval; If the difference of each parameter is within the corresponding deviation threshold interval, no regulation is required. If the difference of each parameter is not within the corresponding deviation threshold interval, regulation is required; Take the parameter that is not within the corresponding deviation threshold interval as the parameter to be adjusted and calculate the difference of the parameter to be adjusted. The difference of the parameter to be adjusted is obtained by the difference of the parameter corresponding to the parameter to be adjusted and the minimum adjustment value of the boundary of the corresponding deviation threshold interval.

6. The intelligent laboratory environment control system according to claim 5, wherein The determination of the control quantities of the corresponding devices through each PID controller corresponding to the difference of the parameters to be adjusted specifically includes: The control quantity of the device corresponding to the parameter to be adjusted is comprehensively calculated through the proportional gain, integral gain, and derivative gain of the parameter to be adjusted to construct the expression of the corresponding PID controller; According to the expression of the PID controller corresponding to the parameter to be adjusted, calculate the control quantity output by the PID controller corresponding to the difference of the parameter to be adjusted to the corresponding device.

7. An intelligent laboratory environment control system according to claim 6, wherein, Encoding the difference of the parameter to be adjusted and the control quantity of the corresponding device into a state vector and inputting it into the state space of reinforcement learning, and selecting the highest reward action based on a preset memory table, using the highest reward action to optimize the control quantity of the corresponding device and performing regulation on the device until there is no difference in the parameter to be adjusted in the current environmental parameters, specifically including: Set a period, and encode the difference of the parameter to be adjusted and the control quantity of the corresponding device into a state vector according to the period and input it into the state space of reinforcement learning; Then, the state vector selects an action from a preset memory table. The action selection includes: retrieving the high-reward action of the highest similar state from the memory table as the action. If there is no matching item in the memory table, use the ε-greedy strategy to explore a new action; Superimpose the adjustment amount of the control quantity of the device corresponding to the selected action on the control quantity output by the PID controller corresponding to the device to form a final control signal and send it to the corresponding device for execution. Calculate the reward value for the action based on the regulation time, regulation times, and safety violation times corresponding to the selected action. Insert the action and state vector corresponding to the reward value into the memory table and update the memory table.

8. An intelligent laboratory environment control system according to claim 7, wherein, The memory table specifically includes: Combine the action corresponding to the state vector and the reward value into an array and store it in the memory table as a new policy. Set an update rule and update the memory table based on the update rule.

9. An intelligent laboratory environment control system according to claim 1, characterized in that, The regulation optimization module further includes an emergency takeover regulation mechanism. The emergency takeover regulation mechanism specifically includes: When any parameter exceeds the corresponding preset safety threshold, switch the device corresponding to the parameter greater than or equal to the preset safety threshold to the maximum allowable regulation mode until the corresponding parameter is less than the preset safety threshold.

10. An intelligent laboratory environment regulation method for implementing an intelligent laboratory environment regulation system according to any one of claims 1 to 9, characterized in that, Including the following steps: Step 1: Construct and train a compensation and correction model through the collected historical environmental parameters and their corresponding historical compensation information, and perform compensation and correction on the currently collected original environmental parameters based on the compensation and correction model to obtain the current environmental parameters; Step 2: Determine the target environmental parameters of the laboratory according to the experimental content in the laboratory; Step 3: Determine the difference of the parameter to be adjusted in the current environmental parameters according to the target environmental parameters, and respectively determine the control quantity of the corresponding device through each PID controller corresponding to the difference of the parameter to be adjusted. Encode the difference of the parameter to be adjusted and the control quantity of the corresponding device into a state vector and input it into the state space of reinforcement learning, and select the highest reward action based on a preset memory table. Use the highest reward action to optimize the control quantity of the corresponding device and perform regulation on the device until there is no difference in the parameter to be adjusted in the current environmental parameters.

Images